The Agony of Opacity: Foundations for Reflective Interpretability in AI-Mediated Mental Health Support

Throughout history, a prevailing paradigm in mental healthcare has been one in which distressed people may receive treatment with little understanding around how their experience is perceived by their care provider, and in turn, the decisions made by their provider around how treatment will progress. Paralleling this offline model of care, people who seek mental health support from AI chatbots are similarly provided little context for how their expressions of distress are processed by the model, and subsequently, the logic that may underlie model responses. People in severe distress who turn to AI chatbots for support thus find themselves caught between black boxes, with unique forms of agony that arise from these intersecting opacities, including misinterpreting model outputs or attributing greater capabilities to a model than are yet possible, which has led to documented real-world harms. Building on empirical research from clinical psychology and AI safety, alongside rights-oriented frameworks from medical ethics, we describe how the distinct psychological state induced by severe distress can influence chatbot interaction patterns, and argue that this state of mind (combined with differences in how a user might perceive a chatbot compared to a care provider) uniquely necessitates a higher standard of interpretability in comparison to general AI chatbot use. Drawing inspiration from newer interpretable treatment paradigms, we then describe specific technical and interface design approaches that could be used to adapt interpretability strategies from four specific mental health fields (psychotherapy, community-based crisis intervention, psychiatry, and care authorization) to AI models, including consideration of the role of interpretability in the treatment process and tensions that may arise with greater interpretability.